v0.1.4: kernel_launcher_impl.h token

[Patrick Kirsch]

Hey,
in kernel_launcher_impl.h:179
There is something mentioned about tokens.
I do not fully understand what that inferes.

As of e.g. CUDA Runtime API v2.2 it (token) is now cudaStream_t:
/usr/local/cuda/include/cuda_runtime_api.h:215
cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem
__dv(0), cudaStream_t stream __dv(0));

Regards,
Patrick

[Jens Breitbart]

As you can see in the comment in the source file. Tokens were not documented in the early CUDA versions. What was called tokens internally in the original version are now the stream identifiers, see for example here:

http://developer.download.nvidia.com/compute/cuda/3_0-Beta1/toolkit/docs/online/group__CUDART__EXECUTION_gcd41b066b3811e07984045e4bc6fc1bb.html

Currently there is a default value being used for token for all kernels.

Best,

Jens